基于CEEMD的LSTM和ARIMA模型干旱预测适用性研究——以新疆为例

doi:10.13866/j.azr.2022.03.07

摘要/Abstract

摘要：

干旱的频繁发生对农业生产和经济发展造成了不可忽视的危害,准确预测干旱的发生具有重要的现实意义。基于1960—2019年新疆气象站点的逐日降水量数据,计算1、3、6、9、12个月及24个月时间尺度的标准化降水指数。建立差分自回归移动平均模型（Autoregressive Integrated Moving Average,ARIMA）、长短期记忆网络（Long Short-Term Memory,LSTM）、互补集合经验模态分解（Complementary Ensemble Empirical Mode Decomposition,CEEMD）-ARIMA组合模型和CEEMD-LSTM组合模型。通过4种模型对多时间尺度SPI序列进行预测,确定各模型在干旱预测中的适用性。结果表明：（1） 4种模型的预测精度均随时间尺度的增加而逐渐提高,在24个月时间尺度时达到最高;（2） CEEMD能够有效平稳时间序列,各时间尺度下,组合模型均达到了较高的预测精度,相较单一模型更适用于干旱预测;（3） 4种模型预测结果精度由低到高分别为：LSTM、ARIMA、CEEMD-LSTM、CEEMD-ARIMA（决定系数最大值分别为：0.8882、0.9103、0.9403、0.9846）,CEEMD-ARIMA模型相比其他3种模型效果较好,最适用于干旱预测。

关键词: 互补集合经验模态分解, 长短期记忆网络, 差分自回归移动平均模型, 标准化降水指数, 干旱预测, 新疆

Abstract:

The frequent occurrence of droughts seriously affects normal agricultural production and economic development. Accurate prediction of drought occurrence is of great importance in reducing drought losses. Nevertheless, drought occurrences have not been well predicted. Drought indices can be used to quantitatively evaluate the intensity, duration, and influence range of drought. Thus, on the basis of daily precipitation data from 1960 to 2019 in the Xinjiang Uyghur Autonomous Region, the standardized precipitation index (SPI) at timescales of 1, 3, 6, 9, 12, and 24 months were calculated. Aiming for the nonlinear and nonstationary characteristics of SPI, a new drought prediction method was proposed combining the single model and the complementary ensemble empirical mode decomposition (CEEMD), which can process nonlinear and nonstationary signals. In this paper, the autoregressive integrated moving average (ARIMA) model, the long short-term memory (LSTM) network, the CEEMD-ARIMA combined model, and the CEEMD-LSTM combined model were constructed to predict a multiscale SPI. The validity of prediction models was determined using root mean square error, mean absolute error, and coefficient of determination (R²). Kriging interpolation was used to demonstrate the predicted results of the four models. The results revealed that the forecast accuracy of the four models increases with the increase of SPI timescales, and the highest accuracy is obtained at SPI24. CEEMD decomposition can effectively stabilize the time series. Drought prediction based on the CEEMD provides a stable premise for the single model. At each timescale, combined models obtain higher prediction accuracy than single models, which indicates that combined models are more suitable for drought prediction. The forecast accuracy of the four models in order from the lowest to highest accuracy is the LSTM model, followed by the ARIMA, CEEMD-LSTM, and CEEMD-ARIMA models (the maximum R² values are 0.8882, 0.9103, 0.9403, and 0.9846, respectively). The CEEMD-ARIMA model shows the best ability to forecast SPI values. This study explored the applicability of four drought prediction models and provided a basis for meteorological disaster prevention and mitigation efforts.

Key words: complementary ensemble empirical mode decomposition, long short-term memory network, autoregressive integrated moving average, standardized precipitation index, drought prediction, Xinjiang

丁严,许德合,曹连海,管相荣. 基于CEEMD的LSTM和ARIMA模型干旱预测适用性研究——以新疆为例[J]. 干旱区研究, 2022, 39(3): 734-744.

DING Yan,XU Dehe,CAO Lianhai,GUAN Xiangrong. Applicability of the LSTM and ARIMA model in drought prediction based on CEEMD: A case study of Xinjiang[J]. Arid Zone Research, 2022, 39(3): 734-744.

图/表 10

图1

表1

图2

图3

图4

表2

表3

图5

表4

图6

参考文献 34

[1]	李夫鹏, 王正涛, 超能芳, 等. 利用Swarm星群探测亚马逊流域2015-2016年干旱事件[J]. 武汉大学学报·信息科学版, 2020, 45(4): 595-603.
	[ Li Fupeng, Wang Zhengtao, Chao Nengfang, et al. 2015-2016 drought event in the Amazon River basin as measured by Swarm constellation[J]. Geomatics and Information Science of Wuhan University, 2020, 45(4): 595-603. ]
[2]	田丰, 武建军, 刘雷震, 等. 1901-2015年华北平原干旱时空转移特征及热点区域探测[J]. 干旱区资源与环境, 2020, 34(6): 87-96.
	[ Tian Feng, Wu Jianjun, Liu Leizhen, et al. Spatiotemporal transferring characteristics of drought and its hotpots detection in North China Plain during 1901-2015[J]. Journal of Arid Land Resources and Environment, 2020, 34(6): 87-96. ]
[3]	卢宝宝, 孙慧兰, 姜泉泉, 等. 近53 a新疆水分盈亏量时空变化特征[J]. 干旱区研究, 2021, 38(6): 1579-1589.
	[ Lu Baobao, Sun Huilan, Jiang Quanquan, et al. Spatiotemporal variation characteristics of the water budget in Xinjiang during the latest 53 years[J]. Arid Zone Research, 2021, 38(6): 1579-1589. ]
[4]	宋玉鑫, 左其亭, 马军霞. 基于SWAT模型的开都河流域水文干旱变化特征及驱动因子分析[J]. 干旱区研究, 2021, 38(3): 610-617.
	[ Song Yuxin, Zuo Qiting, Ma Junxia. Variation and dynamic drivers of drought in Kaidu River Basin based on the SWAT model[J]. Arid Zone Research, 2021, 38(3): 610-617. ]
[5]	Vasiliades L, Loukas A, Liberis N. A water balance derived drought index for Pinios River Basin, Greece[J]. Water Resources Management, 2011, 25(4): 1087-1101. doi: 10.1007/s11269-010-9665-1
[6]	方秀琴, 郭晓萌, 袁玲, 等. 随机森林算法在全球干旱评估中的应用[J]. 地球信息科学学报, 2021, 23(6): 1040-1049. doi: 10.12082/dqxxkx.2021.200474
	[ Fang Xiuqin, Guo Xiaomeng, Yuan Ling, et al. Application of random forest algorithm in global drought assessment[J]. Journal of Geo-Information Science, 2021, 23(6): 1040-1049. ] doi: 10.12082/dqxxkx.2021.200474
[7]	刘媛媛, 李霞, 王小博, 等. 2001-2018年中国-老挝交通走廊核心区植被稳定性对极端干旱的响应[J]. 生态学报, 2021, 41(7): 2537-2547.
	[ Liu Yuanyuan, Li Xia, Wang Xiaobo, et al. Vegetation stability in response to extreme droughts from 2001 to 2018 in the core area of China-Laos transportation corridors[J]. Acta Ecologica Sinica, 2021, 41(7): 2537-2547. ]
[8]	周丽, 谢舒蕾, 吴彬. 基于CI和强度分析方法的四川冬春季干旱事件变化特征[J]. 自然灾害学报, 2020, 29(3): 36-44.
	[ Zhou Li, Xie Shulei, Wu Bin. Variation characteristics of the winter and spring drought events in Sichuan based on CI and Intensity analysis[J]. Journal of Natural Disasters, 2020, 29(3): 36-44. ]
[9]	徐一丹, 任传友, 马熙达, 等. 基于SPI/SPEI指数的东北地区多时间尺度干旱变化特征对比分析[J]. 干旱区研究, 2017, 34(6): 1250-1262.
	[ Xu Yidan, Ren Chuanyou, Ma Xida, et al. Change of drought at multiple temporal scales based on SPI /SPEI in Northeast China[J]. Arid Zone Research, 2017, 34(6): 1250-1262. ]
[10]	李明, 葛晨昊, 邓宇莹, 等. 黄土高原气象干旱和农业干旱特征及其相互关系研究[J]. 地理科学, 2020, 40(12): 2105-2114. doi: 10.13249/j.cnki.sgs.2020.12.017
	[ Li Ming, Ge Chenhao, Deng Yuying, et al. Meteorological and agricultural drought characteristics and their relationship across the Loess Plateau[J]. Scientia Geographica Sinica, 2020, 40(12): 2105-2114. ] doi: 10.13249/j.cnki.sgs.2020.12.017
[11]	Rezaei H, Faaljou H, Mansourfar G. Stock price prediction using deep learning and frequency decomposition[J]. Expert Systems with Applications, 2020, 169(12): 114332, doi: 10.1016/j.eswa.2020.114332. doi: 10.1016/j.eswa.2020.114332
[12]	徐岩岩, 常军. 基于DERF2.0模式1-52天最低温度逐日预报的检验评估[J]. 高原气象, 2018, 37(4): 1042-1050. doi: 10.7522/j.issn.1000-0534.2017.00090
	[ Xu Yanyan, Chang Jun. Evaluation of the minimum temperature forecast of 1-52 days based on DERF2.0 model[J]. Plateau Meteorology, 2018, 37(4): 1042-1050. ] doi: 10.7522/j.issn.1000-0534.2017.00090
[13]	Wu C, Wang J, Chen X, et al. A novel hybrid system based on multi-objective optimization for wind speed forecasting[J]. Renewable Energy, 2020, 146(8): 149-165. doi: 10.1016/j.renene.2019.04.157
[14]	王蕾, 王鹏新, 李俐, 等. 应用条件植被温度指数预测县域尺度小麦单产[J]. 武汉大学学报·信息科学版, 2018, 43(10): 1566-1573.
	[ Wang Lei, Wang Pengxin, et al. Wheat yield forecasting at county scale based on time series vegetation temperature condition index[J]. Geomatics and Information Science of Wuhan University, 2018, 43(10): 1566-1573. ]
[15]	杨慧荣, 张玉虎, 崔恒建, 等. ARIMA和ANN模型的干旱预测适用性研究[J]. 干旱区地理, 2018, 41(5): 945-953.
	[ Yang Huirong, Zhang Yuhu, Cui Hengjian, et al. Applicability of ARIMA and ANN models for drought forecasting[J]. Arid Land Geography, 2018, 41(5): 945-953. ]
[16]	Liu M D, Ding L, Bai Y L. Application of hybrid model based on empirical mode decomposition, novel recurrent neural networks and the ARIMA to wind speed prediction[J]. Energy Conversion and Management, 2021, 233: 113917, doi: 10.1016/j.enconman.2021.113917. doi: 10.1016/j.enconman.2021.113917
[17]	张建海, 张棋, 许德合, 等. ARIMA-LSTM组合模型在基于SPI干旱预测中的应用--以青海省为例[J]. 干旱区地理, 2020, 43(4): 1004-1013.
	[ Zhang Jianhai, Zhang Qi, Xu Dehe, et al. Application of a combined ARIMA-LSTM model based on SPI for the forecast of drought: A case study in Qinghai Province[J]. Arid Land Geography, 2020, 43(4): 1004-1013. ]
[18]	王亦斌, 孙涛, 梁雪春, 等. 基于EMD-LSTM模型的河流水量水位预测[J]. 水利水电科技进展, 2020, 40(6): 40-47.
	[ Wang Yibin, Sun Tao, Liang Xuechun, et al. Prediction of river water flow and water level based on EMD-LSTM model[J]. Advances in Science and Technology of Water Resources, 2020, 40(6): 40-47. ]
[19]	刘艳, 杨耘, 聂磊, 等. 玛纳斯河出山口径流EEMD-ARIMA预测[J]. 水土保持研究, 2017, 24(6): 273-280, 285.
	[ Liu Yan, Yang Yun, Nie Lei, et al. The EEMD-ARIMA prediction of runoff at mountain pass of Manas River[J]. Research of Soil and Water Conservation, 2017, 24(6): 273-280, 285. ]
[20]	杨倩, 秦莉, 高培, 等. 基于EEMD-LSTM模型的天山北坡经济带年降水量预测[J]. 干旱区研究, 2021, 38(5): 1235-1243.
	[ Yang Qian, Qin Li, Gao Pei, et al. Prediction of annual precipitation in the northern slope economic belt of Tianshan Mountains based on a EEMD-LSTM model[J]. Arid Zone Research, 2021, 38(5): 1235-1243. ]
[21]	Zhang X, Wu X, He S, et al. Precipitation forecast based on CEEMD-LSTM coupled model[J]. Water Science & Technology Water Supply, 2021, 21(22): 1-17.
[22]	许德合, 丁严, 张棋, 等. EEMD-ARIMA在干旱预测中的应用--以新疆维吾尔自治区为例[J]. 中国农村水利水电, 2021(7): 1-11.
	[ Xu Dehe, Ding Yan, Zhang Qi, et al. Application of the EEMD-ARIMA combined model in drought prediction: A case study in Xinjiang Uygur Autonomous Region[J]. China Rural Water and Hydropower, 2021(7): 1-11. ]
[23]	Tsakiris G, Vangelis H. Towards a drought watch system based on spatial SPI[J]. Water Resources Management, 2004, 18(1): 1-12. doi: 10.1023/B:WARM.0000015410.47014.a4
[24]	Moreira E E. SPI drought class prediction using log-linear models applied to wet and dry seasons[J]. Physics and Chemistry of the Earth, 2016, 94: 136-145.
[25]	Alquraish M, Abuhasel K A, Alqahtani A S,et al. SPI-based hybrid hidden Markov-GA, ARIMA-GA, and ARIMA-GA-ANN models for meteorological drought forecasting[J]. Sustainability, 2021, 13(22): 12576, doi: 10.3390/su132212576. doi: 10.3390/su132212576
[26]	Huang N E, Shen Z, Long S R, et al. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis[J]. Proceedings Mathematical Physical & Engineering Sciences, 1998, 454(1971): 903-995.
[27]	Wu Z, Huang N E. Ensemble empirical mode decomposition: A noise-assisted data analysis method[J]. Advances in Adaptive Data Analysis, 2009, 1: 1-41. doi: 10.1142/S1793536909000047
[28]	Yeh J R, Shieh J S, Huang N E. Complementary ensemble empirical mode decomposition: A novel noise enhanced data analysis method[J]. Advances in Adaptive Data Analysis, 2010, 2(2): 135-156. doi: 10.1142/S1793536910000422
[29]	孙堃, 赵萌萌, 沈美娜, 等. 基于CEEMD和模糊熵的随机森林风力发电功率预测[J]. 智慧电力, 2019, 47(10): 36-43.
	[ Sun Kun, Zhao Mengmeng, Shen Meina, et al. Wind power forecasting with random forest based on CEEMD and fuzzy entropy[J]. Smart Power, 2019, 47(10): 36-43. ]
[30]	Box G E P, Jenkins G M. Time series Analysis: Forecasting and Control[M]. San Francisco: Holden Day, 1976.
[31]	Dilling S, Macvicar B J. Cleaning high-frequency velocity profile data with autoregressive moving average (ARMA) models[J]. Flow Measurement & Instrumentation, 2017, 54: 68-81.
[32]	左秀霞. 带高次趋势项的ADF单位根检验[J]. 数量经济技术经济研究, 2019, 36(1): 152-169.
	[ Zuo Xiuxia. ADF unit rot test with high order trend term[J]. The Journal of Quantitative & Technical Economics, 2019, 36(1): 152-169. ]
[33]	Liu Q, Zhang G, Ali S, et al. SPI-based drought simulation and prediction using ARMA-GARCH model[J]. Applied Mathematics and Computation, 2019, 355: 96-107. doi: 10.1016/j.amc.2019.02.058
[34]	Adikari K E, Shrestha S, Ratnayake D T, et al. Evaluation of artificial intelligence models for flood and drought forecasting in arid and tropical regions[J]. Environmental Modelling & Software, 2021, 144(4): 105136, doi: 10.1016/j.envsoft.2021.105136. doi: 10.1016/j.envsoft.2021.105136

SPI范围	类型
SPI>-0.5	无旱
-1.0<SPI≤-0.5	轻旱
-1.5<SPI≤-1.0	中旱
-2.0<SPI≤-1.5	重旱
SPI≤-2.0	特旱

SPI序列	单位根检验	临界值			P值
SPI序列	单位根检验	1%	5%	10%	P值
SPI1	-8.0407	-3.4419	-2.8667	-2.5695	1.8500e-12
SPI3	-9.4801	-3.4419	-2.8666	-2.5695	3.8938e-16
SPI6	-6.7711	-3.4420	-2.8667	-2.5695	2.6407e-09
SPI9	-4.4529	-3.4423	-2.8668	-2.5696	0.0002
SPI12	-4.0259	-3.4423	-2.8668	-2.5696	0.0013
SPI24	-3.8011	-3.4425	-2.8669	-2.5696	0.0029

SPI序列	p	q	AIC	BIC
SPI1	1	0	1752.561	1766.295
SPI3	0	2	1511.110	1529.410
SPI6	4	1	1274.699	1306.696
SPI9	2	1	1044.266	1067.099
SPI12	2	1	648.904	671.716
SPI24	9	2	181.161	240.251

时间尺度	模型	R²	RMSE	MAE
1个月	LSTM	-0.0146	0.8681	0.6478
	ARIMA	-0.0058	0.8643	0.6431
	CEEMD-LSTM	0.2648	0.7389	0.5683
	CEEMD-ARIMA	0.4488	0.6398	0.4828
3个月	LSTM	0.4200	0.7906	0.6040
	ARIMA	0.4986	0.7350	0.5531
	CEEMD-LSTM	0.5782	0.6742	0.5017
	CEEMD-ARIMA	0.8246	0.4347	0.3355
6个月	LSTM	0.6686	0.6595	0.4710
	ARIMA	0.6870	0.6410	0.4554
	CEEMD-LSTM	0.7776	0.5402	0.4116
	CEEMD-ARIMA	0.9153	0.3334	0.2397
9个月	LSTM	0.7873	0.5732	0.3856
	ARIMA	0.8039	0.5503	0.3553
	CEEMD-LSTM	0.8921	0.4082	0.2839
	CEEMD-ARIMA	0.9619	0.2426	0.1789
12个月	LSTM	0.8592	0.4858	0.3084
	ARIMA	0.8732	0.4610	0.2628
	CEEMD-LSTM	0.9302	0.3420	0.2251
	CEEMD-ARIMA	0.9793	0.1863	0.1271
24个月	LSTM	0.8882	0.4266	0.2700
	ARIMA	0.9103	0.3822	0.2109
	CEEMD-LSTM	0.9403	0.3119	0.1958
	CEEMD-ARIMA	0.9846	0.1584	0.1019